Construction and Building Materials 49 (2013) 672–681

Contents lists available at ScienceDirect

Construction and Building Materials

journal homepage: www.elsevier.com/locate/conbuildmat

Effect of volumetric factors on the mechanical behavior of asphalt fine

aggregate matrix and the relationship to asphalt mixture properties
B. Shane Underwood ⇑,1, Y. Richard Kim
North Carolina State University, Department of Civil, Construction, and Environmental Engineering, Campus Box 7908, Raleigh, NC 27695-7908, USA

h i g h l i g h t s

 The properties of FAM are evaluated with respect to asphalt and air void content changes.
 Quantitative comparisons between FAM and AC properties with compositional changes are made.
 A micromechanical model is used to study the linkages between the FAM and AC moduli.
 An empirical correlation is used to study the linkages between the FAM and AC fatigue properties.
 FAM can be useful for both practical and modeling tasks with proper material design and testing.

a r t i c l e i n f o a b s t r a c t

Article history: The multiscale study of asphalt concrete using fine aggregate matrix (FAM) has become widespread in
Received 21 February 2013 recent years. Different laboratory fabrication procedures have been proposed, and since FAM’s sensitivity
Received in revised form 2 August 2013 to compositional effects is unknown this literature cannot be coherently interpreted. In this paper, the
Accepted 29 August 2013
mechanical responses of FAM at different volumetric compositions are systematically studied. The visco-
Available online 25 September 2013
elastic and tensile properties are found to be sensitive to volumetric composition. It is concluded that the
use of FAM for modeling purposes requires accurate replication of FAM as it exists in the mixture. How-
Keywords:
ever, such strict requirements are not necessary for simpler, comparative evaluations.
Asphalt concrete
Fine aggregate matrix
Ó 2013 Elsevier Ltd. All rights reserved.
Multiscale evaluation
Linear viscoelastic
Fatigue

1. Introduction
Asphalt fine aggregate matrix (FAM) consists of asphalt binder,
filler-sized particles, fine aggregate particles, and air and it exists in
the interstitial spaces between the coarse aggregate particles with-
in an asphalt concrete (AC) mixture. FAM is important because it is
a single characteristic length scale smaller than the AC and is
therefore closer in characteristic size of the damage that occurs
within the AC [1]. Experiments with FAM have been used to study
fatigue damage, moisture damage and healing in AC mixtures, with
the argument that the phenomena occur largely between the
coarsest aggregate particles and so tests with FAM should provide
direct indications of how they will affect AC mixture [1–4]. FAM
materials are characteristically similar to AC mixtures with the pri-
mary difference being that the maximum aggregate size is gener-
⇑ Corresponding author. Tel.: +1 480 965 1097; fax: +1 480 965 0557.
E-mail address: Shane.Underwood@asu.edu (B.S. Underwood).
1
Arizona State University, School of Sustainability and the Built Environment, PO
Box 875306, Tempe, AZ 85287-5306, USA.
ally smaller than 2.36 mm and so test samples can be created at
smaller geometries and still meet representative volume element,
RVE, requirements [5]. Maintaining the RVE requirements ensures
that measured properties are not functions of the test geometry
and that they represent the fundamental characteristics of the
material. Being able to fabricate and test small geometries means
that mechanistically viable experiments can be carried out using
less costly equipment and in less time than would be required
for larger geometries [6,7].
The question remains whether the benefits of FAM experiments
outweigh the limitations of needing to utilize upscaling models to
predict the properties of the AC. It is the contention of this paper
that the benefits do outweigh these limitations, but that appropri-
ate care must be taken in both the experimental phase, and the
upscaling process. Both computational [8–12] and analytical [13–
22] based upscaling approaches have been attempted with varying
levels of success. The primary limitation with these approaches
thus far has been the lack of an experimentally verified, functional
based microstructural hypothesis for supporting the numerical dis-
cretization. These works primarily focus on applications to the case

0950-0618/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.conbuildmat.2013.08.045
 B.S. Underwood, Y.R. Kim / Construction and Building Materials 49 (2013) 672–681 673

of non-damaged composites, but some have been expanded to ex-
plore the effects of localized disturbances on macromechanical re-
sponses [23–25]. The overall conclusion from these modeling
efforts is that in order to completely understand the macrome-
chanical response of a heterogeneous composite, one must first
understand the mechanical response of that same composite at dif-
ferent length scales.
One shortcoming in the available literature is the detailed
understanding of the fundamental characteristics of FAM as a func-
tion of volumetric composition changes. This lack of understanding
makes it difficult to view the literature as a coherent set of infor-
mation since most researchers utilize their own techniques to de-
fine and create FAM. One pragmatic approach is to set the largest
size of aggregate in FAM based on resolution limitations with dig-
ital image analysis (2.36–2 mm) [26,27]. Kim and Aragao [28] pro-
posed that the FAM contained particles smaller than 1.18 mm and
all of the non-absorbed asphalt cement. Recently, Sousa et al. [29]
also suggested the 1.18 mm threshold, but determined the asphalt
content by staged sieve analysis. The authors of this paper have
also proposed a method based on a series of microstructural and
gravimetric experiments [30]. This method theorizes that all aggre-
gates in the mixture are coated with a film of asphalt mastic (filler
particles and asphalt) and that the FAM exists between the coated
coarse aggregate particles. The aggregate size in the FAM is defined
according to the nominal maximum aggregate size of the mixture
evaluations of the material properties with respect to sensitivity
to volumetric changes. Such an evaluation can benefit analytical/
computational modeling efforts, but can also be useful for evaluat-
ing the effectiveness of FAM-based experiments in practical engi-
neering applications like; performance based mix design,
selecting amongst material alternatives for a given application,
and/or rapid forensic studies [32–33]. The objectives of this paper
are to demonstrate and quantify;
 How changes in asphalt and air void content affect the
dynamic shear modulus, |G|, and tensile properties
(strength and strain at failure) of FAM, and
 How FAM and AC mixture behaviors are related through
analytical micromechanical modeling for |G| and through
correlation for damaged properties.
The unique contribution of this research is in the systematic
evaluation of these effects and in the quantification of composi-
tional effects on modulus and tensile properties. These findings
have implications in both engineering practice and multiscale
modeling applications and will be useful in reviewing and compil-
ing conclusions from the existing literature.
2. Materials and test method

and the asphalt content is determined based on the film of mastic
coating the FAM aggregates. The FAM is further estimated to con-
tain 40–70% of the air within the AC mixture based on analysis of
computed tomography images of AC mixture. [30]. Aside from vol-
umetric composition, compaction of the FAM is also an issue and
Izadi et al. found that gyratory compaction replicated the micro-
structure of FAM as it exists in the mixture [31]. Obviously, there
is no firm consensus on how the FAM exists in the mixture and
how to replicate this in the laboratory. The purpose of this paper
is not to rank order the potential fabrication methods. Rather,
the data presented in this paper provides insight into why it is
important to identify a method to replicate the FAM, and under
what specific applications a more or less accurate replication is
needed.
While some of the previous research has evaluated FAM charac-
teristics for select conditions, none have carried out systematic
2.1. Materials
Two granite-based AC mixtures with highly angular, cubical, low deleterious
content, and well graded aggregate are used for this study. The first mixture has
a nominal maximum size of aggregate (NMSA) of 9.5 mm and the second has a
NMSA of 19.0 mm. These materials are respectively referred to as the 9.5 mm and
19.0 mm mixtures throughout this paper, and were chosen to provide different
sized representative FAM as discussed below. The FAM materials are labeled as
either coarse FAM (C-FAM) or very fine FAM (VF-FAM), depending on the maximum
aggregate size. The C-FAM consists of aggregates passing the 2.36 mm sieve. The
maximum aggregate size for the VF-FAM is based on the experimental findings
from the authors that the maximum aggregate size in the FAM is equal to the fine
aggregate initial break sieve used in the Bailey method of mix design [30,34];
0.6 mm for the 9.5 mm mixture and 1.18 mm for the 19 mm mixture. Different
FAM materials have been created at different asphalt binder and air void contents
within these maximum aggregate sizes as shown in Table 1 and Fig. 1.
With respect to the materials listed in Table 1, the asphalt content was con-
trolled by carefully weighing in a target mass of asphalt binder during fabrication.
Although the mean air void contents are shown in Table 1 it is noted that all of the

Table 1
Test conditions and designations for the primary mixtures.

Mix FAM Size Name Asphalt content by mass (%) Air void (%) Direct tensiona
9.5 mm C-FAM CFL-9 8.27 9.1
CFM-9 9.70 9.1
CFM-7 9.70 6.5
CFM-5 9.70 4.5
CFH-9 11.2 9.1
CFH-7 11.2 6.5
VF-FAM VFAS-4 16.5 3.3 X
VFBS-4 15.2 3.9 X
VFBS-8 15.2 7.8 X
VFCS-4 13.2 4.5 X
VFCS-8 13.2 8.0 X
VFDH-12 11.6 12.4 X
VFDH-19 11.6 19.4 X
VFDL-21 8.5 21.0
VFEH-4 15.6 4.4 X
VFEL-10 12.2 9.5
VFFS-2 16.3 1.7
VFFS-5 16.3 4.6
19.0 mm VF-FAM VF19AS-2 13.1 1.9
VF19AS-4 13.1 3.7
VF19BS-5 11.7 5.0 X
VF19BS-8 11.7 8.2 X
a
X indicates that direct tension testing was performed.
674 B.S. Underwood, Y.R. Kim / Construction and Building Materials 49 (2013) 672–681
100
0.075
80
(a)
% Passing
% Passing
60
40
20
C-FAM
0 0
0.15 0.3 0.6 1.18 2.36
Sieve Size (mm)
Fig. 1. Compilation of gradations for study FAM materials: (a) C-FAM and (b) VF-FAM.
replicates were within ±0.5% of the reported mean. Air void content was deter-
mined by comparing the measured bulk specific gravity (measured by AASTHO
T-166) with the maximum specific gravity (measured by AASHTO T-209). The
T-166 test samples were 12 mm in diameter and 45 mm tall and so a balance with
0.01 g readability was used to ensure a valid measurement of sufficient precision.
The process of fabricating test samples follows the basic technique developed by
Zollinger [4].
Multiple VF-FAM gradations were evaluated and in the naming convention, the
third letter denotes the particle gradation with each letter (A, B, C, D, E, and F) rep-
resenting a different gradation, the letters only serve to distinguishing the materials
and have no other special meaning. In the case of the C-FAM, the third letter de-
notes the asphalt content. The fourth letter in the VF-FAM materials indicates the
relative asphalt content for that gradation (L = Low, H = High, and S = only a single
asphalt content was evaluated). Finally, the number that appears after the dash is
the target air voids content (the actual as-measured air void content is given in
Table 1). Of the FAM shown in Table 1, VFBS-8 and VF19BS-8 are of particular note
since they follow from the authors hypothesis on microstructural assemblage are
therefore believed, by the authors, to best represent the FAM as it exists within
the AC mixture.
2.2. Test methods
2.2.1. Temperature and frequency sweep tests
Temperature and frequency sweep tests were conducted on the FAM materials
to determine the linear viscoelastic (LVE) characteristics using a TA Instruments AR-
G2 rheometer. FAM samples were glued to plates machined to match the testing
fixture. Unless otherwise noted the strain magnitude used for each test condition
was 60 le, and approximately 5–20 cycles were applied at each frequency (14,
6.5, 3, 1.4, 0.65, 0.3, 0.14, and 0.1 Hz). For the C-FAM test temperatures are 14°,
23°, 38°, and 55 °C and for the VF-FAM the temperatures are 10°, 20°, 35°, and
54 °C. The primary response functions of interest are the |G| and phase angle, d.
2.2.2. Direct tension tests
Tensile properties were evaluated with direct tension tests. The FAM materials
that were tested in this manner are indicated with an X-mark in the last column of
Table 1. Tension samples were first glued to end plates and then linear variable dif-
ferential transducers were attached to the samples surface to measure on-specimen
deformation. Then, a closed-loop servo-hydraulic loading machine applied a con-
stant rate of deformation to the test sample. Due to machine and test set-up com-
pliance issues, the observed on-specimen strain rate was not constant, but this
shortcoming was overcome by analyzing the data by using the measured on-spec-
imen deformation. Tests were performed until the specimens completely separated
at 10 °C and at rates between 0.015 and 0.6902 e/s.
3. Experimental results
3.1. Temperature and frequency sweep tests
The temperature and frequency sweep data is analyzed based
on an assumption of thermorheological simplicity. The data is
shifted horizontally to construct mastercurves use well established
techniques that are detailed elsewhere [35]. The amount of tem-
perature dependent horizontal shift needed to create a master-
curve is the time–temperature shift factor (t–T shift factor). The
reduced frequency, xR, is the product of the test frequency, x,
and the t–T shift factor for the test temperature, aT, xR = x  aT.
The typical LVE characterization curves (i.e., |G| mastercurve, d
mastercurve, and t–T shift factor function) for FAM materials are
100
0.075
0.075
80
(b)
60 VFA
VFB
VFC
40 VFD
VFE
20 VFF
VF19A
VF19B
0.15 0.3 0.6 1.18
Sieve Size (mm)
shown along with the AC mixture curves in Fig. 2. Many of the
same overall characteristics of AC mixtures also occur in FAM
materials: (1) S-shaped mastercurve function, (2) increasing then
decreasing phase angle, and (3) a similar t–T shift factor function.
As is shown later in this paper, these qualitative similarities exist
for all FAM materials, but the specific shape and magnitude of
the LVE response functions vary. The qualitative similarities be-
tween FAM and mix are not a particularly new finding. However,
detailed quantitative similarities and differences are discussed in
more detail later in this paper.
3.1.1. Effect of asphalt content
In AC, increases in asphalt content cause a reduction in the
modulus value and an increase in the phase angle. Experimental
studies suggest that the effects range from 10% to 55% change in
modulus per percentage change in asphalt content [32,36]. That
is, changing the asphalt content from 4% to 5% may reduce the
modulus value by 10–55% depending upon the specific material
and temperature/frequency of interest. This particular indicator
for sensitivity is chosen for its convenience and familiarity to engi-
neers. Typically changes in asphalt content (and air void content)
are expressed not as percent changes, but rather absolute changes
in the percentage by total volume. Thus, in practice a change in
asphalt content from 4% to 5% is not viewed as a 20% change in
asphalt content (although it is), but rather as a 1 percentage change
in asphalt content. Fig. 3 shows the percent change in |G| per per-
centage of change in asphalt content as a function of temperature
for the C-FAM. Using the available database, it is most convenient
to examine the effects of asphalt content for the C-FAM; however,
similar observations can be made for the VF-FAM materials after
accounting for air void effects. The data in Fig. 3 show that the
effects of asphalt content change on the FAM |G| are similar in
magnitude to the effects of asphalt content change on the AC mod-
ulus. The data in this figure was gathered at 18.85 rad/s. Compari-
sons at other frequencies result in the same conclusions, but
different values for the sensitivity indicator.
In Fig. 3, mixture data from a separate study is shown [32].
These mixtures consist of aggregates with similar gradation, struc-
ture, geographic origin, and asphalt type as in the current investi-
gation. Although the sensitivity is not exactly the same, the
magnitude of these effects is similar between FAM and AC, approx-
imately 20–35% change in modulus per percentage change in as-
phalt content. Both mixture and FAM show that the sensitivity is
maximized at approximately 35 °C. This finding is not surprising
since this is the region of the mastercurve where frequency depen-
dence is greatest and therefore a gross change in modulus or to the
time-dependence will affect these results the most. There are
physical implications of this similarity. Using volumetric calcula-
tions along with the aforementioned definition of FAM, it can be
shown that since a larger amount of the asphalt cement exists in
the FAM phase, a single percentage change in the asphalt content
 B.S. Underwood, Y.R. Kim / Construction and Building Materials 49 (2013) 672–681 675

1.E+10 6.E+09
(a) 5.E+09
(b)
4.E+09

|G*| (Pa)
|G*| (Pa)
1.E+08 3.E+09

2.E+09

1.E+09 Mix
Mix
FAM FAM
1.E+06 0.E+00
1.E-05 1.E-03 1.E-01 1.E+01 1.E+03 1.E+05 1.E-05 1.E-03 1.E-01 1.E+01 1.E+03 1.E+05
Reduced Frequency (rad/s) Reduced Frequency (rad/s)

60 2

50
(c) 1 (d)
Phase Angle (deg)

Log Shift Factor

0
40
-1
30
-2
20
-3
10 Mix -4 Mix
FAM FAM
0 -5
1.E-05 1.E-03 1.E-01 1.E+01 1.E+03 0 20 40 60
Reduced Frequency (rad/s) Temperature (°C)

Fig. 2. Comparison of characteristic LVE behaviors of FAM materials and AC mixtures: (a) |G| in logarithmic space, (b) |G| in semi-logarithmic space, (c) phase angle, and (d)
t–T shift factor.

35
9.5 mm mixture shows approximately a 5% change in modulus
per percentage change in air void content. It is interesting that this
(% Change in |G*|) / (Δ % AC)

30 sensitivity is not dependent on the FAM NMSA, which may suggest

that the overall internal structure assembles in a self-similar way
25
with scale. One slight outlier to this observation is the VFDH com-
20 parison (i.e., VFDH-12 versus VFDH-19). The sensitivity indicator
for VFDH is computed at a higher air void content than the others,
15 and this finding may be related to an increased sensitivity to air
void change at extremely high air void contents. Observing the
10
CFAM FAM from the 19.0 mm mixture, the sensitivity to air void content
5 Lcontrol - Lee (2007) changes is approximately double what is observed in the 9.5 mm
Lsub - Lee (2007) based FAM. Comparisons made with AC (MX9.5 and MX19) of sim-
0 ilar physical and manufactured characteristics as the study FAM
0 10 20 30 40 50 60
Temperature (°C) show that, like the FAM experiments, the 19.0 mm materials have
a greater sensitivity to air void content. Therefore, it is promising
Fig. 3. Effects of asphalt content change on FAM |G| (evaluated at 18.85 rad/s). that the FAM tests show this trend in behavior.

of the AC will equate to approximately a three percentage change

in asphalt content of the FAM [4,28–30]. Since the sensitivity to a 14
(% Change in |G*|) / ( Δ % AV)

single percentage change in asphalt content is similar, this condi-

tion means that FAM experiments may be up to three times as 12
capable of detecting the impacts of asphalt content changes on 10
AC modulus.
8

6
3.1.2. Effect of air void content
Air void content also affects the modulus of AC mixtures, and it 4
has been experimentally observed than an increase in the air void
2
content causes a mostly vertical shift in the mastercurve when
plotted in the logarithmic domain [37]. In FAM, a similar pattern 0
is observed, which is quantified by calculating the percent change
in the modulus versus the percentage change in air void content, as
shown in Fig. 4. The results are found to be similar across test tem-
9.5 mm Materials 19.0 mm Materials
peratures and frequencies (consistent with the vertical shift obser-
vation), and so, for convenience, only the values from the Fig. 4. Effects of air void content change on the |G| of FAM materials, and
18.85 rad/s frequency tested at 20 °C are shown. FAM from the comparisons of this sensitivity with sensitivities of similar AC mixtures.
676 B.S. Underwood, Y.R. Kim / Construction and Building Materials 49 (2013) 672–681
The results also support the hypothesis from the introduction
that much of the mechanical and engineering characteristics of
AC mixtures are impacted by the characteristic behaviors of the
FAM material. It remains to be shown that the FAM responses
are sensitive enough to distinguish between materials with similar
characteristics, i.e., between two different 9.5 mm mixtures. It
should be noted that like the case of the asphalt content, a single
percentage change in the air voids of the mixture translates to a
larger percentage change in air void content of the FAM. The exact
correspondence is dependent upon the make-up of the material,
specifically the volumetric contribution of the FAM to the whole
asphalt concrete mixture and the amount of the total air in the
mixture that is held in the FAM. For typical dense graded mixtures,
volumetric calculations can be used to show that this difference is
between approximately 1.8 and 1.2 (i.e., a 1% change in the air void
content of a mixture will cause a 1.8–1.2% change in the air void
content of the FAM). Therefore, the similarity in sensitivity to a sin-
gle percentage change means that FAM based evaluation will be
more capable of detecting the impact of air void content changes
on mixture behavior.
3.1.3. Effect of FAM fabrication parameters
Thus far it has been shown that FAM shows similarities to AC
mixture with respect to its sensitivity to percentage changes in as-
phalt and air void content. Such a finding implies that engineering
decisions based on comparisons amongst alternatives can be made
without having to exactly replicate FAM as it exists in the mixture.
However, FAM experiments may be capable of providing much
more than qualitative comparisons, and it is believed that experi-
mental results can be used for more advanced modeling and anal-
ysis. Such analysis requires accurate absolute characterization of
the FAM materials as opposed to accurate relative characterization.
Selected |G| mastercurves from the 9.5 mm FAM materials are
shown in Fig. 5 to demonstrate the overall impact of fabrication
method. The modulus values are found to vary by as much as
50–80% depending on the exact fabrication method and tempera-
ture/frequency of interest. Although not shown, a similar variation
with the 19.0 mm based FAM materials (20–70%) is also observed.
The data in Fig. 5 indicates that the FAM fabrication protocol can-
not be chosen based simply on convenience and then be expected
to yield useful mixture-level insights, because such an approach
may not consider the requisite model’s physical assumptions. Sim-
ilarly, accurate material-level understanding cannot be gained
using a complex analytical or computational model that considers
physical mechanisms that may or may not actually exist in the
material. The material models and fabrication protocols must be
developed together so that they are consistent with each other
and with the conditions that exist in the AC mixture. It is the
1.E+10
|G*| (Pa)
1.E+08
1.E+06
1.E-06 1.E-04 1.E-02 1.E+00
Reduced Frequency (rad/s)
Fig. 5. Comparison of |G| mastercurves for selected FAM materials.
authors’ opinion that this consistency rule is imperative for ensur-
ing that the outcomes from multiscale modeling efforts are accu-
rate and provide useful material insights. It should also be
recognized that Fig. 5 represents only the LVE effects and that as
other mechanisms are activated, greater differences may occur.
3.2. Direct tension
The relationship between the tensile properties (strength and
strain tolerance) and fatigue performance of AC mixtures is been
well established [38–41]. In this study the tensile properties are
thus examined to provide surrogate insight into the fatigue behav-
iors of these FAM materials. The strength and strain tolerances are
categorized to show the effects of air void and asphalt content in
Figs. 6 and 7 respectively. In Fig. 6, comparisons with the high
and low void content VFCS and VFBS FAM show that the strength
can reduce by 10% with a single percentage increase in air void
content. This effect is less pronounced for the VFDH materials,
which were tested at overall higher air void contents than the VFCS
and VFBS materials. At these air void contents, a percentage change
in air void level may not constitute as much of a change in actual
air voids in the system since a percentage of change in air void con-
tent at 4% air void constitutes a 25% change in air void, whereas at a
12% air void content the effect constitutes only an 8.3% change. The
failure strains are found to be less sensitive to air void content
changes, generally showing an increase of 0–5% per percentage
of increase in air void content.
The volumetric composition of VFAS, VFBS, and VFCS and so
comparisons between these materials highlight the effects of as-
phalt content change. For these comparisons, only the materials
with air void contents of approximately 4% were examined, and
it is found that a percentage increase in asphalt content reduces
the strength by 4–6% and increases the strain tolerance by 25–
40%. One possible reason for the difference in sensitivity is that
film strain may be the critical factor affecting the failure of FAM.
This film would be thicker when more asphalt binder is included
in the FAM, the consequence of which is that the localized strain
that exists under a prescribed global strain would be smaller than
when a relatively thinner film existed. Thus, the materials with
higher asphalt contents could resist greater global strain before
failing. The reduced strength would also follow from this hypothe-
sis because the film strain rate would be lower at the same overall
strain rate, and, as Figs. 6 and 7 show, viscoelastic materials are
weaker at smaller strain rates. This finding does have implications
with respect to mixture behaviors since the empirical relationships
between axial tension and fatigue cracking are based strain at fail-
ure. AC fatigue performance shows an approximately 30–45%
improvement in fatigue life per percentage of change in asphalt
content [42]. This level of sensitivity is similar to what is observed
in Fig. 7(b).
4. FAM to mixture upscaling
A better fundamental understanding of the upscaling mechan-
ics is needed in order to evaluate the use of FAM in more exacting
CFL-9
tasks. Here, the assembly is viewed as a two-phase composite con-
CFH-9
sisting of coarse aggregate particles and FAM. Although local vari-
VFAS-4
ations in stress and strain states are known to exist, it is assumed
VFDL-21
that over the volume of a mixture sample (typically 100 mm diam-
VFFS-2
eter  150 mm tall) that the local variations average to a stable
1.E+02 1.E+04 condition. With guidance from reported values, it is further as-
sumed that this average condition is equivalent to a simple strain
ratio where the local FAM strain is seven times greater than the
global strain [12,43,44].
 B.S. Underwood, Y.R. Kim / Construction and Building Materials 49 (2013) 672–681 677

5000 15000
(a) (b) VFBS-4 VFBS-8

με)
Peak Stress (kPa)

Strain at Failure (με

4000 12000 VFCS-4 VFCS-8
VFDH-12 VFDH-19

3000 9000

2000 6000
VFBS-4 VFBS-8
1000 VFCS-4 VFCS-8
3000
VFDH-12 VFDH-19
0 0
1.E-04 1.E-03 1.E-02 1.E-01 1.E-04 1.E-03 1.E-02 1.E-01
Strain Rate (ε/s) Strain Rate (ε/s)

Fig. 6. Direct tension test results showing the effects of air void content: (a) strength and (b) strain at failure.

5000 15000
VFAS-4
(a) (b)

με)
Strain at Failure (με
Peak Stress (kPa)

VFBS-4
4000 12000
VFCS-4
VFEH-4
3000 9000 VFFS-5

2000 VFAS-4 6000

VFBS-4
1000 VFCS-4 3000
VFEH-4
VFFS-5
0 0
1.E-04 1.E-03 1.E-02 1.E-01 1.E-04 1.E-03 1.E-02 1.E-01
Strain Rate (εε/s) Strain Rate (ε/s)

Fig. 7. Direct tension test results showing the effects of asphalt content: (a) strength and (b) strain at failure.

Many analytical models exist for explaining this two-phase
scenario including the Mori–Tanaka scheme [15], the General-
ized-Self Consistent method (GSCM) [16,17], the composite
sphere model [18], the self-consistent model [19], the periodic
structures method [20], and the differential scheme [21,22].
Although each of these formulations is basically similar, they dif-
fer slightly in either the assumptions used to derive the formulas
or in the mathematical steps followed. Thus, it is not expected
that they would produce exactly the same nor drastically different
predictions. Nevertheless, the Christensen and Lo GSCM model
[16] is found to provide the best solution. This model assumes
spherical shapes, isotropic elasticity, and strain energy equiva-
lence between the idealized microstructure and the equivalent
medium.
Christensen and Lo use elastic theory to show that the general
elastic solutions (Ur and Uh) in the mixture, M, FAM, F, and partic-
ulate, P, phases are given by Eq. (1) where m and |m| denotes the
Poisson’s Ratio and dynamic Poisson’s ratio and the subscript de-
notes the phase with which the quantity is associated.
ð1Þ
Continuity of stresses and displacements at the interfaces sug-
gests that eight different inequalities must be satisfied, and each
of these contains a combination of the coefficients A1 through D4
and phase specific Poisson’s ratios and shear moduli. To these con-
tinuity conditions, simple shear boundary conditions are applied
under the requirement that the strain energy must be conserved
between the model problem and the real configuration. Ultimately,
these constraints lead to the quadratic function given in the follow-
ing equation:
  2   
jG jM jG jM
A  þB þC ¼0 ð2Þ
jG jF jG jF
The outcome of this solution is the ratio of the mixture modu-
lus, |G|M, to the FAM modulus, |G|F. The functions A, B, and C are
dependent upon the material properties and content of the FAM
and aggregate particles, and require some lengthy calculations that
are summarized in Eq. (4). In these formulas GP is the elastic shear
modulus of the aggregate, mP is the Poisson’s ratio of the aggregate,
|G|F is the dynamic shear modulus of the FAM, |m|F is the dynamic
Poisson’s ratio of the FAM, Cv is the volumetric particle concentra-
tion, and j1 through j3 are used to simplify the presentation of
functions A, B, and C.
Comparisons of the measured |G|M and that predicted from Eq.
(2) are given in Fig. 8(a and b) for the study mixtures. Results from
the Mori–Tanaka scheme, Fig. 8(c and d), and the differential
scheme, Fig. 8(e and f), are also shown. Note, the composite sphere
model is also popular, but is not shown here since it provides for a
bounded solution of |G|M and here the interest is in an explicit pre-
diction of the quantity. The properties of VFBS-8 and VF19BS-8 are
used as input for these functions because, as indicated earlier, the
authors believe them to be the most representative of the FAM that
exists within the AC. Experiments with the VFBS-8 FAM at 54 °C
could not be completed due to its relatively low stiffness at these
conditions and the limitations of the available equipment. It should
678 B.S. Underwood, Y.R. Kim / Construction and Building Materials 49 (2013) 672–681

1.E+10 1.E+10
R2 = 0.98 R2 = 0.96

1.E+09 1.E+09

|G*| (Pa)
|G*| (Pa)
1.E+08 1.E+08

1.E+07 1.E+07
20°C-Pred. 20°C-Meas. 20°C-Pred. 40°C-Pred. 54°C-Pred.

1.E+06
(a) 40°C-Pred. 40°C-Meas.
1.E+06
(b) 20°C-Meas. 40°C-Meas. 54°C-Meas.

1.E-01 1.E+00 1.E+01 1.E+02 1.E+03 1.E-01 1.E+00 1.E+01 1.E+02 1.E+03
Frequency (rad/s) Frequency (rad/s)

1.E+10 1.E+10
R2 = 0.83 R2= 0.80

1.E+09 1.E+09

|G*| (Pa)
|G*| (Pa)

1.E+08 1.E+08

1.E+07 1.E+07
20°C-Pred. 20°C-Meas. 20°C-Pred. 40°C-Pred. 54°C-Pred.

1.E+06
(c) 40°C-Pred. 40°C-Meas.
1.E+06
(d) 20°C-Meas. 40°C-Meas. 54°C-Meas.

1.E-01 1.E+00 1.E+01 1.E+02 1.E+03 1.E-01 1.E+00 1.E+01 1.E+02 1.E+03
Frequency (rad/s) Frequency (rad/s)

1.E+10 1.E+10
R2= 0.95 R2= 0.96

1.E+09 1.E+09
|G*| (Pa)

|G*| (Pa)

1.E+08 1.E+08

1.E+07 1.E+07
20°C-Pred. 20°C-Meas. 20°C-Pred. 40°C-Pred. 54°C-Pred.

1.E+06
(e) 40°C-Pred. 40°C-Meas.
1.E+06
(f) 20°C-Meas. 40°C-Meas. 54°C-Meas.

1.E-01 1.E+00 1.E+01 1.E+02 1.E+03 1.E-01 1.E+00 1.E+01 1.E+02 1.E+03
Frequency (rad/s) Frequency (rad/s)

Fig. 8. Comparison of measured and predicted mixture modulus for; (a) 9.5 mm mixture with GSCM, (b) 19.0 mm mixture with GSCM, (c) 9.5 mm mixture with composite
sphere model, (d) 19.0 mm mixture with composite sphere model, (e) 9.5 mm mixture with differential scheme model, and (f) 19.0 mm mixture with differential scheme
model.

1.E+05 1.E+03
(a) (b)
με)

Strain at Nf = 1 x 106
Strain at Failure (με

1.E+04 1.E+02

VFBS-8 9.5 mm Mix

VF19BS-8 19.0 mm Mix
1.E+03 1.E+01
1.E-04 1.E-03 1.E-02 1.E-01 1.E-01 1.E+00 1.E+01 1.E+02 1.E+03
Strain Rate (ε/s) Reduced Frequency (rad/s)

Fig. 9. (a) Comparison of 9.5 mm and 19.0 mm FAM strain tolerance and (b) comparison of fatigue behavior of 9.5 mm and 19.0 mm mixtures.

also be noted that the mixture modulus was measured via axial
experiment, but both axial and radial deformation were recorded.
From these quantities the temperature and frequency dependent
Poisson’s ratio, |m|M, was calculated (|m|M varied from approxi-
mately 0.25 at 10 °C and 14 Hz to 0.47 at 54 °C and 0.1 Hz) and
used to convert the axial dynamic modulus, |E|M, to |G|M by
|G|M = |E|M/(2(1 + |m|M)). Although the Poisson’s ratio of FAM
was not measured, it was assumed that |m|F = |m|M for all temper-
ature and frequencies. This assumption seemed reasonable given
the overall qualitative and quantitative similarity between the
moduli of the mixture and FAM. Additionally, some sensitivity
analysis was conducted with different values of Poisson’s ratio
and the overall trends for each model remained the same. Since
mixture experiments were conducted at peak-to-peak values of
 B.S. Underwood, Y.R. Kim / Construction and Building Materials 49 (2013) 672–681 679

60 le, the FAM data used in this analysis was gathered at peak-to- 0.5 the physical system transitions from being one where the mod-
peak values of 420 le. ulus is dependent on the inverse of (1  Cv) to one where it is
  dependent on the inverse of (1  Cv)3 [45]. Christensen shows that
GP
A¼8  1 ð4  5jm jF Þj1 C 10=3
v the GSCM reflects this trend, whereas the differential scheme sug-
jG jF gests that the modulus varies with (1  Cv)2 in the compressible
     
GP 7=3 GP case and (1  Cv)5/2 when Poisson’s ratio is 0.5. Since Poisson’s ratio
 2 63  1 j 2 þ 2j j
1 3 C v þ 252  1 j2 C 5=3
v
jG jF jG jF increases with temperature (for the same frequency) it is expected
  that the differential scheme will tend to predict higher values than
GP
 50  1 ð7  12jm jF þ 8jm j2F Þj2 C v þ 4ð7 the GSCM, which is supported by the 10 °C predictions. As it so
jG jF
happens within the Poisson’s value ranges encountered in the
 10jm jF Þj2 j3 experiments the two models predict similar values, but based on
  the limit analysis and based on the fact that mixture modulus val-
GP ues are sometimes needed at temperatures as low as 10 °C where
B ¼ 4  1 ð1  5jm jF Þj1 C 10=3
v
jG jF the Poisson’s ratio will likely decrease further, the GSCM is a better
     
GP 7=3 GF physical representation of the FAM to mixture upscaling process.
þ 4 63  1 j 2 þ 2 j j
1 3 C v  504  1 j2 C 5=3
v An interesting observation about the agreement shown in
jG jF jG jF
  Fig. 8(a and b), and to a lesser extent in parts (e and f), is that it
GP
þ 150  1 ð3  jm jF Þm jF j2 C v þ 3ð15jm jF  7Þj2 j3 is substantially better than some other computational based
jG jF
upscaling methods [47,48]. The question arises as to why a simple
  effective medium based analytical model can accurately predict
Gp  10=3 the upscaling of this complex structure. This question is particu-
C¼4   1 ð5jm jF  7Þj1 C v
jG jF larly relevant since the GSCM does not explicitly account for parti-
     
Gp 7=3 Gp 5=3 cle size distribution, or the shape and angularity of the coarse
 2 63   1 j2 þ 2j1 j3 C v þ 252   1 j2 C v
jG jF jG jF aggregates (it does implicitly account for these factors in the fine
  aggregate). The idealization of a dense graded asphalt concrete
Gp  2 
þ 25   1 ðjm jF  7Þj2 C v  ð7 þ 5jm jF Þj2 j3 mixture as a two phase composite consisting of only FAM and
jG jF
coarse aggregates is still one that contains approximately 50–60%
    coarse aggregate particles. Thus, the modeled particulate inclu-
GP  GP
j1 ¼  1 ð49  50mP jm j Þ þ 35 ðmP  2jm jF Þ sions are likely still interacting within this idealized volume. How-
jG jF F
jG jF ever, it should be kept in mind that the GSCM is based on global
þ 35ð2mP  jm jF Þ averages of stress, strain, and energy. So while locally these quan-
     
G G GP tities might be affected by the coarse aggregate parameters, glob-
j2 ¼ 5mp P  8 þ 7 P þ 4 ; j3 ¼  ð8  10jm jF Þ ally they are likely more dependent on the bulk properties and
jG jF jG jF jG jF
volumetric concentrations of the constituent phases [46]. This is
þ ð7  5jm jF Þ
particularly true when the matrix and composite share a similar
ð3Þ modulus magnitude and would explain why effective medium
In Fig. 8 the three models produce similar results; however, the models are incapable of bridging the gap between asphalt cement
Mori–Tanaka model is found to be noticeably less accurate than binder and asphalt concrete where the difference in modulus may
the other two. Christensen [45] explains how this method is phys- be several orders of magnitude. The reliance on volumetric content
ically unrealistic for high concentrations and an incompressible and modulus of the matrix phase in this upscaling framework cou-
matrix. This conclusion seems reasonable based on the experimen- pled with the differences observed in the previous sections of this
tal data here since upscaling involves relatively high particle con- paper is also a strong argument for why careful attention to the
centration (approximately 60% by volume) and the errors FAM composition must be given in order to obtain the results
increase as the FAM Poisson’s ratio moves towards 0.5 (e.g., at shown in Fig. 8. When damage occurs, localized variables take on
higher temperature). The differences between the differential an increased emphasis and therefore it is not expected that the
scheme and GSCM are less obvious, but the computed coefficient GSCM will be capable of bridging the gap between FAM and mix-
of determination, R2 from the GSCM predictions is smaller than ture once the LVE range is exceeded.
or equal to that from the differential scheme. The R2 is defined These findings also suggest an interesting alternative hypothe-
using Eq. (5) where the numerator of the function is the sum of sis for the role and applicable scales of internal structure. Classi-
squared errors and the denominator is the total sum of squares cally, aggregate structure is considered in the context of the
in the logarithmic domain. In this equations |G|M,pred is the value skeleton formed by the larger (4.75 mm and above) aggregate par-
predicted from a model, |G|M,meas is the measured modulus, ticles. Since the GSCM does not explicitly account for particle con-
log(|G|M)avg is the mean of the measured logarithm of |G| at all tact or higher order interactions, these effects, if they exist, must be
test conditions, the index i represents the particular temperature accounted for in the properties of the matrix phase. Since larger
and frequency, and N is the total number of temperatures and particles are not part of the FAM and the upscaling predictions
frequencies. are generally positive, it may be inferred that within the asphalt
mixture many of these interactions occur at the FAM and lower
PN h
    i2
scales. In the analysis considered here, the level of deformation is
log jG jM;pred  log jG jM;meas
i¼1
R2 ¼ 1  i i
ð4Þ relatively small (60 le). At greater levels of strain the large particle
PN h 



i2
i¼1 log jG jM;meas  log ðjG jM Þav g skeleton affects may be more substantial. At the very least then,
i
this observation leads to the hypothesis that the internal structural
The differential model and GSCM need not predict the same or factors contributing to the observed engineering behaviors of a
similar values. However, it occurs in these cases for two reasons; mixture vary with length scale and deformation level interactively.
(1) the particle concentration is relatively high and (2) for the con- The consequence of this theory is that mixture decisions (changes
ditions evaluated the Poisson’s ratio varies only between 0.25 and to gradation for example) should be made with particular behav-
0.47. At high particle concentrations, as Poisson’s ratio approaches iors and deformations in mind.
680 B.S. Underwood, Y.R. Kim / Construction and Building Materials 49 (2013) 672–681
To show that the direct tension results relate to the fatigue
behavior of AC mixtures, the data for FAM with VFBS-8 and
VF19BS-8 are combined with fatigue analysis results from the
9.5 mm and 19.0 mm mixtures. Unlike the comparisons shown in
the |G| sections of this paper, these comparisons use AC mixtures
from the same aggregate and asphalt source. Fig. 9(a) compares the
failure strain results from the direct tension tests on the FAM
materials and, based on this graph alone, the FAM from the
9.5 mm mix performs better than the FAM obtained from the
19.0 mm mix. If these results are generalized based on the empir-
ical relationship proposed by other researchers [31,32], then it can
be surmised that the strain at failure in direct tension testing cor-
relates closely to the strain level that yields fatigue failure at one
million cycles. This relationship is reflected by first analyzing the
mixture data to extract the same information (i.e., strain level for
1 million cycles to failure) at 10 Hz loading and at different tem-
peratures (5°, 19°, and 27 °C) according to the procedure detailed
elsewhere [49]. These predictions are compiled into a domain sim-
ilar to that used to show the monotonic test data (y-axis of strain
and x-axis of strain rate) by finding the reduced frequency, for each
prediction, and then plotting this value against the strain level for
one million cycles to failure, as shown in Fig. 9(b).
Using this analysis, both the FAM and AC tests indicate that; (1)
the 9.5 mm materials perform better than the 19 mm materials
and (2) the difference in performance changes depending on the
rate, frequency, temperature, etc. The fact that both experiments
capture this trend is limited evidence that FAM-based studies
can separate the performance characteristics of different AC mix-
tures. Of course, further study is needed to determine whether this
method is capable of accurately detecting smaller variations in
mixture composition, but the results shown in Fig. 9 are positive
enough to warrant such study.
5. Conclusions
This study finds that the LVE and tensile properties of FAM
materials are affected by the volumetric composition of the FAM
materials. For the study materials, the specific findings include:
 A one percentage increase in asphalt content results in a 20–
35% decrease in |G|, a 4–6% decrease in strength, and a 25–
40% increase in strain tolerance.
 A one percentage increase in air void content results in a 5–8%
decrease in |G| for the 9.5 mm FAM, a 10–12% decrease in |G|
for the 19.0 mm FAM, a 10% decrease in strength, and a 0–5%
increase in strain tolerance.
These ranges in sensitivity combined with the more general
observation regarding similarities between AC and FAM leads to
the following two conclusions regarding fabrication of FAM for
the purposes of experiment: when used as qualitative screening
or in comparative studies it is not necessary to replicate FAM ex-
actly as it exists in the mixture as long as consistent rules are fol-
lowed in fabricating each study material, but when used for
quantitative analysis and/or modeling purposes one must care-
fully consider the composition of the FAM as it exists within
the asphalt concrete mixture. Based on the experimental data in
this paper, failure to carefully consider FAM composition may
mean that the insight gathered from the modeling efforts could
be flawed. Several hypotheses exist regarding FAM composition,
but additional work is needed to verify these hypotheses and de-
velop consensus. Once a valid hypothesis exists, accurate meth-
ods to mechanistically describe the link between the scales can
be developed.
References
[1] Kim YR. Mechanistic fatigue characterization and damage modeling of asphalt
mixtures. Doctoral Dissertation; Texas A&M University; 2003.
[2] Castelo Branco VTF, Masad E, Bhasin A, Little DN. Fatigue analysis of asphalt
mixtures independent of mode of loading. Transp Res Rec 2008;2057:149–56.
[3] Palvadi S, Bhasin A, Little DN. A method to quantify healing in asphalt
composites using continuum damage approach. Transp Res Rec
2011;2296:86–96.
[4] Zollinger J. Application of surface energy measurements to evaluate moisture
susceptibility of asphalt and aggregates. Master’s thesis, Texas A&M
University; 2005.
[5] Hashin Z. Analysis of composite materials – a survey. J Appl Mech
1983;29:143–50.
[6] Goodrich JL. Asphalt binder rheology, asphalt concrete rheology and asphalt
concrete mix properties. J Assoc Asphalt Paving Technol 1991;60:80–120.
[7] Velasquez R, Zofka A, Turos M, Marasteanu MO. Bending beam rheometer
testing of asphalt mixtures. Int J Pave Eng 2011;12:461–74.
[8] You Z, Adhikari S, Dai Q. Air void effect on an idealized asphalt mixture using
two-dimensional and three-dimensional discrete element modeling approach.
Int J Pave Eng 2010;11:381–91.
[9] Aragao FTS, Kim YR, Lee J, Allen DH. Micromechanical model for heterogeneous
asphalt concrete mixtures subjected to fracture failure. J Mater Civ Eng
2011;23:30–8.
[10] Masad E, Somadevan N. Microstructural finite-element analysis of influence of
localized strain distribution on asphalt mix properties. J Eng Mech
2002;128:1105–14.
[11] Wang H, Hao P. Numerical simulation of indirect test based on the
microstructure of asphalt mixture. J Mater Civ Eng 2011;23:21–9.
[12] Kim YR, Souza FV, Teixeira JESL. A two-way coupled multiscale model for
predicting damage-associated performance of asphaltic roadways. Comput
Mech 2013;51:187–201.
[13] Shenoy A. Rheology of filled polymer systems. Norwell (MA): Kluwer
Academic Publishers; 1999.
[14] Aragao FTS, Kim YR, Karki P, Little DN. Semiempirical, analytical, and
computational predictions of dynamic modulus of asphalt concrete
mixtures. Transp Res Rec 2010;2181:19–27.
[15] Benveniste Y. A new approach to the application of Mori-Tanaka’s theory in
composite materials. Mech Mater 1987;6:147–57.
[16] Christensen RM, Lo KH. Solutions for effective shear properties in three phase
sphere and cylinders models. J Mech Phys Solids 1979;27:315–29.
[17] Halpin JC, Kardos JL. The Halpin-Tsai equations: a review. Polym Eng Sci
1976;16:344–52.
[18] Hashin Z. The elastic moduli of heterogeneous materials. J Appl Mech
1962;29:143–50.
[19] Hill R. A self-consistent mechanics of composite materials. J Mech Phys Solids
1965;13:213–22.
[20] Yin HM, Sun LZ. Elastic modelling of periodic composites with particle
interactions. Philos Mag Lett 2005;85:163–73.
[21] Kim M, Buttlar WG. Differential scheme effective medium theory for hot-mix
asphalt |E| prediction. J Mater Civ Eng 2011;23:69–78.
[22] McLaughlin R. A study of the differential scheme for composite materials. Int J
Eng Sci 1977;15:237–44.
[23] Bazant ZP, Tabbara MP, Kazemi MT, Cabot GP. Random particle model for
fracture of aggregate or fiber composites. J Eng Mech 1990;116:1686–705.
[24] Schlangen E, VanMier JGM. Experimental and numerical analysis of
micromechanisms of fracture of cement-based composites. Cem Concr
Compos 1992;14:105–18.
[25] Huet C. Coupled size and boundary-condition effects in viscoelastic
heterogeneous and composite bodies. Mech Mater 1999;31:787–829.
[26] Kim H, Wagoner MP, Buttlar WG. Simulation of fracture behavior in asphalt
concrete using a heterogeneous cohesive zone discrete element model. J Mater
Civ Eng 2008;20:552–63.
[27] Valenta R, Sejnoha M, Zeman J. Macroscopic constitutive law for mastic
asphalt mixtures from multiscale modeling. J Multiscale Comput Eng
2010;8:131–49.
[28] Kim YR, Aragao FTS. Microstructure modeling of rate-dependent fracture
behavior in bituminous paving mixtures. Finite Elem Anal Des 2013;63:23–32.
[29] Sousa P, Kassem E, Masad E, Little DN. New design method of fine aggregate
mixtures and automated method for analysis of dynamic mechanical
characterization data. Constr Build Mater 2013;41:216–23.
[30] Underwood BS, Kim YR. Microstructural investigation of asphalt concrete for
performing multiscale experimental studies. Int J Pave Eng 2012;14:498–516.
[31] Izadi A, Motamed A, Bhasin A. Designing fine aggregate mixtures to evaluate
fatigue crack-growth in asphalt mixtures. Final report, SWUTC/11/161022-1,
Southwest Region University Transportation Center, Texas Transportation
Institute; 2011.
[32] Lee S. Investigation of the effect of lime on performance of HMA using
advanced testing and modeling techniques. Doctoral Dissertation; North
Carolina State University; 2007.
[33] Zhou F, Hu S, Scullion T, Chen DH, Qi X, Claros G. Development and verification
of the overlay tester based fatigue cracking prediction approach. J Assoc
Asphalt Paving Technol 2007;76:627–62.
[34] Vavrik WR, Pine WJ, Huber G, Carpenter SH, Bailey R. The bailey method of
gradation evaluation: the influence of aggregate gradation and packing
 B.S. Underwood, Y.R. Kim / Construction and Building Materials 49 (2013) 672–681
characteristics on voids in the mineral aggregate. J Assoc Asphalt Paving
Technol 2001;70:132–75.
[35] Underwood BS, Kim YR. Experimental investigation into the multiscale
behavior of asphalt concrete. Int J Pave Eng 2011;12:357–70.
[36] Shook JF, Kallas BF. Factors influencing dynamic modulus of asphalt concrete. J
Assoc Asphalt Paving Technol 1969;38:140–78.
[37] Kim YR, Lee S, Seo Y, El-Haggan O. Impact of price reductions on the long-term
pavement performance of HMA mixes in North Carolina. Final report, FHWA/
NC/2005-09, North Carolina Department of Transportation; 2005.
[38] Bonnot J. Asphalt aggregate mixtures. Transp Res Rec 1986;1096:42–50.
[39] Rao Tangella SCS, Craus J, Deacon JA, Monismith CL. Summary report on
fatigue response of asphalt mixtures. Technical report SHRP report A-312,
strategic highway research program. National Research Council, Washington
(DC), USA; 1990.
[40] Epps JA. Influence of mixture variables on the flexural fatigue and tensile
properties of asphalt concrete. Doctoral Dissertation: University of California,
Berkeley; 1969.
[41] Salam YM, Monismith CL. Fracture characteristics of asphalt concrete. J Assoc
Asphalt Paving Technol 1972;41:215–56.
[42] Harvey JT, Tsai BW. Effects of asphalt content and air void content on mix
fatigue and stiffness. Transp Res Rec 1996;1543:38–45.
681
[43] Masad E, Somadevan N, Bahia HU, Kose S. Modeling and experimental
measurements of strain distribution in asphalt mixes. J Transp Eng 2001;127:
477–85.
[44] Seo Y. A Comprehensive study of crack growth in asphalt concrete using
fracture mechanics. Doctoral Dissertation: North Carolina State University;
2003.
[45] Christensen RM. A critical evaluation for a class of micro-mechanics models. J
Mech Phys Solids 1990;38:379–404.
[46] Christensen RM. Mechanics of composite materials. New York (NY): John
Wiley & Sons; 1979.
[47] Adhikari S, You Z. Investigating the sensitivity of aggregate size within sand
mastic by modeling the microstructure of an asphalt mixture. J Mater Civ Eng
2011;23:580–6.
[48] Western Research Institute. Asphalt research consortium: quarterly research
report April 1–June 30, 2010. Federal Highway Administration. Contract No.
DTFH61-07-H-00009; 2010.
[49] Hou T, Underwood BS, Kim YR. Fatigue performance prediction of north
Carolina mixtures using the simplified viscoelastic continuum damage model.
J Assoc Asphalt Paving Technol 2010;79:35–80.